Adversarially Robust Abductive Fusion of Pre-trained Transformer-based Perception Models

1Key Takeaways

• arXiv:2608.04190v1 Announce Type: new Abstract: Deploying pre-trained perception models in novel environments degrades their accuracy under distributional shift, and assembling them alone does not recover it: combiners such as majority voting trade recall for precision and are brittle to coordinated failures.
• Prior metacognitive methods learn logical rules that flag a model's errors, but rely on hand-authored domain-knowledge cues (object-size priors, segmentation masks) that do not transfer to….